Skip to content
AI Chief of Staff — AI/ML

AI Chief of Staff

Jan – Apr 2026

PythonLangGraphFastAPIPostgreSQLRedisECS Fargate

consulting engagements go wrong quietly — a stalled Linear board, a channel that went silent, a PR nobody reviewed. this ingests all of it from Notion, Slack, Linear and GitHub and turns it into a prioritised risk briefing. i owned the event-driven core: the part that has to keep one project's events in order while several projects are being analysed at once.

designed the event-driven core of a 7-engineer system scoring delivery risk across concurrent client projects from 32 webhook types

3-phase ingest worker: per-project Redis locks and timestamp-scored queues order each project's events while projects analyse concurrently

retry backoff and dead-letter queues, because webhook payloads are sparse and every event needs a follow-up API call that can fail

what i built

  • the ingest worker: HMAC-verified webhooks in, Redis queue, ordered per-project processing out
  • per-project locks plus timestamp-scored queues so ordering holds within a project without serialising across projects
  • retry backoff and dead-letter handling for the follow-up fetches that sparse webhooks force

how it works

  1. 1FastAPI validates the HMAC signature, extracts metadata, enqueues and returns 200 immediately
  2. 2webhook payloads are sparse, so a worker fetches the actual content from each platform's API
  3. 3a per-project lock and a timestamp-scored queue keep that project's events in order
  4. 4agents score the signals and the dashboard surfaces prioritised risk

results

  • shipped as the ingest backbone for a 7-engineer build across 32 webhook types